How can I find the specific code implementation involved in KV calculation and writing the calculation results into cache?

[YnRen22852]

1. I browsed vllm/attention/backends/abstract.py and found no underlying code to implement KV calculations, only parameter calls. I want to understand how to do the underlying implementation.
2. I observed relevant code to implement KV calculations in the csrc/attention folder. What is the difference between this and the abstract implementation of vllm above? Which one is the code that actually implements the underlying calculations?